mango

关于LUT的问题

1

结合论文、代码，我的理解是LUT0、LUT1、LUT2是可学习的参数，整个模型最终的目的就是优化目标函数获取一个较好的LUT(LUT=w0*LUT0+w1*LUT1+w2*LUT2)；请问可以把LUT0、LUT1、LUT2看作是可学习的参数吗？

适用LLaMA-Factory微调qwen-7b-chat，所有参数与finetune.py一致，为什么效果差距很大呢？可以从哪些方面分析呢？

3

### Reminder - [X] I have read the README and searched the existing issues. ### Reproduction LLama-factory微调参数：deepspeed --num_gpus 2 src/train_bash.py --deepspeed ./examples/sft/ds_config_zero2_new.json --stage sft --do_train --model_name_or_path './qwen/Qwen-7B-Chat' --dataset luxun_alpace --finetuning_type...

The training completes the first stage，when tested on REDS data set, abnormal color appears, GT:  SR image:  please,in first stage,have you ever been in the same situation?

about VideoLQ dataset

1

In VideoLQ datasets, there are only low-quality LQ images，none HR images. How can it be used as a test?

In paper, "To demonstrate the effects, we compare BasicVSR (using bidirectional propagation) with its unidirectional variant (with comparable network complexity)."please, is the complexity of the network structure of the comparison...

qwen-7b-chat、qwen1.5-7b-chat微调效果对比

9

qwen1.5-7b-chat微调： deepspeed --num_gpus 2 src/train_bash.py --deepspeed ./Qwen1.5/examples/sft/ds_config_zero2_new.json --stage sft --do_train --model_name_or_path './Qwen1.5-7B-Chat' --dataset huxijin_luxun_alpace --finetuning_type lora --lora_target q_proj,k_proj,v_proj,o_proj,up_proj,gate_proj,down_proj --output_dir qwen1.5_7b_chat_huxijin_luxun --overwrite_cache --num_train_epochs 60 --per_device_train_batch_size 2 --gradient_accumulation_steps 8 --save_strategy epoch --lr_scheduler_type...

[https://github.com/QwenLM/Qwen1.5/blob/main/examples/sft/finetune.py#L156](url) 对训练数据处理这块，为什么input_ids、target_ids除了补充的token是不一致的，其他都是一致的呢？按照这个逻辑训练没问题嘛？我理解的模型训练，input_ids应该是 { "role": "user","content": ""}、target_ids应该是{ "role": "assistant", "content": ""}，我的理解有问题嘛？

test code: CUDA_VISIBLE_DEVICES=1 bash ./tools/dist_test.sh configs/TTVSR_reds4.py checkpoints/TTVSR_REDS.pth 1 --save-path 'save_path_REDS' , num_input_frames=10; the output image has serious artifacts,are the weights provided optimal? 

请问单GPU如何训练、测试？

4

把python分布式命令 python -m torch.distributed.launch --nproc_per_node=$GPUS --master_port=$PORT \ $(dirname "$0")/train.py $CONFIG --launcher pytorch ${@:3} 改成 python ./tools/train.py ./configs/FTVSR_reds4.py 运行模型训练报错：KeyError: 'TTVSR is not in the model registry' 请问是什么原因呢？